Traffic Sign Recognition Classifier¶
Project: Build a Traffic Sign Recognition Classifier¶
Traffic signs are used to regulate,warn,guide or inform road users. They can help achieve an acceptable level of road traffic quality and increase safety with orderly and predictable movement of all traffic, both vehicular and pedestrians.
Traffic sign recognition(TSR) whichs act as an essential part of the Advanced Driving Assistance Systems (ADAS) play an important role to gudie and alert the drivers in potentially dangerous situations, for example when a driver may be overspeeding.Traffic sign recognition have a number of difficult challenges due to real-world variabilities such as viewpoint variations, lighting conditions (saturations, low-contrast), motion-blur, occlusions, sun glare, physical damage, colors fading, graffiti, stickers.
In this project,we will build a traffic sign recognition classifier to distinguish raffic signs into lables as correct as possible .This project will design,train and test on German Traffic Sign Dataset.
The following tools are used:
- tensorflow
- opencv
- python
The following libraries are used:
- pickle
- numpy
- pandas
- matplotlib
- random
- skimage
Those following techniques are used:
- Convolutional Neural Network(CNN)
Step 0: Load The Data¶
In [1]:
# Load pickled data
import pickle
import os
import numpy as np
from pandas.io.parsers import read_csv
# TODO: Fill this in based on where you saved the training and testing data
training_file = "traffic-signs-data/train.p"
validation_file = "traffic-signs-data/valid.p"
testing_file = "traffic-signs-data/test.p"
with open(training_file, mode='rb') as f:
train = pickle.load(f)
with open(validation_file, mode='rb') as f:
valid = pickle.load(f)
with open(testing_file, mode='rb') as f:
test = pickle.load(f)
#traffic sign label names stored in signnames.csv
signnames = read_csv("signnames.csv").values[:, 1]
X_train, y_train = train['features'], train['labels']
X_valid, y_valid = valid['features'], valid['labels']
X_test, y_test = test['features'], test['labels']
Step 1: Dataset Summary & Exploration¶
The pickled data is a dictionary with 4 key/value pairs:
'features'is a 4D array containing raw pixel data of the traffic sign images, (num examples, width, height, channels).'labels'is a 1D array containing the label/class id of the traffic sign. The filesignnames.csvcontains id -> name mappings for each id.'sizes'is a list containing tuples, (width, height) representing the original width and height the image.'coords'is a list containing tuples, (x1, y1, x2, y2) representing coordinates of a bounding box around the sign in the image. THESE COORDINATES ASSUME THE ORIGINAL IMAGE. THE PICKLED DATA CONTAINS RESIZED VERSIONS (32 by 32) OF THESE IMAGES
Complete the basic data summary below. Use python, numpy and/or pandas methods to calculate the data summary rather than hard coding the results. For example, the pandas shape method might be useful for calculating some of the summary results.
Provide a Basic Summary of the Data Set Using Python, Numpy and/or Pandas¶
In [2]:
### Replace each question mark with the appropriate value.
### Use python, pandas or numpy methods rather than hard coding the results
import pandas as pd
# TODO: Number of training examples
n_train = len(X_train)
# TODO: Number of validation examples
n_validation = len(X_valid)
# TODO: Number of testing examples.
n_test = len(X_test)
# TODO: What's the shape of an traffic sign image?
image_shape = X_train[0].shape
# TODO: How many unique classes/labels there are in the dataset.
n_classes = pd.unique(pd.Series(y_train)).size
print("Number of training examples =", n_train)
print("Number of validing examples =", n_validation)
print("Number of testing examples =", n_test)
print("Image data shape =", image_shape)
print("Number of classes =", n_classes)
Include an exploratory visualization of the dataset¶
Visualize the German Traffic Signs Dataset using the pickled file(s). This is open ended, suggestions include: plotting traffic sign images, plotting the count of each sign, etc.
Define visualization of function¶
In [3]:
### Data exploration visualization code goes here.
### Feel free to use as many code cells as needed.
import matplotlib.pyplot as plt
import random
# Visualizations will be shown in the notebook.
%matplotlib inline
def show_one_class_images(images, lables, oneclass,defualtshowsize = 10, withexmaple = True,cols = 10,cmap=None):
""""
Random show one specified[oneclass] class image.
Arguments:
images: source images
lables: source corresponding label
oneclass: class id[0-42]
defualtshowsize:maximum showing images per process
cols:image showing column size
"""
if oneclass >= n_classes:
print("Out of range,maximum index range is %d" % (n_classes-1))
return
totshowimage = images[lables == oneclass]
totlsumclass = len(totshowimage)
#show images less than defualtshowsize
if totlsumclass > defualtshowsize:
sumclass = defualtshowsize
else:
sumclass = totlsumclass
rows = (sumclass+cols-1)//cols
print("Show Class=[%d],Name=[%s] from data set,Show length is [%d],Total length is [%d]" % (oneclass,signnames[oneclass],sumclass,totlsumclass))
exampleimage = plt.imread("resources/signs/%d.jpg"%(oneclass))
plt.figure(figsize=(30, 20))
for i in range(sumclass):
plt.subplot(rows, cols, i+1)
if i==0 and withexmaple:
showimage = exampleimage
else:
showimage = totshowimage[np.random.randint(len(totshowimage))]
showimage_shape = showimage.shape
# use gray scale color map if there is only one channel
if len(showimage_shape) == 2:
cmap = "gray"
elif showimage_shape[2] == 1:
showimage = showimage[:,:,0]
cmap = "gray"
plt.imshow(showimage,cmap=cmap)
plt.title(oneclass,fontsize=30,fontweight='bold')
plt.xticks([])
plt.yticks([])
plt.tight_layout(pad=0, h_pad=0, w_pad=0)
plt.show()
def show_all_class_images(images, lables, sumclass=n_classes, cols = 10,cmap=None):
""""
Random show one image of all class from dataset.
Arguments:
images: source images
lables: source corresponding label
sumclass: sumary class(42)
"""
rows = (sumclass+cols-1)//cols
print("Random show each class from data set:")
plt.figure(figsize=(40, 30))
for i in range(sumclass):
plt.subplot(rows, cols, i+1)
labelmarkimg = images[lables == i]
#random show one image
showimage = labelmarkimg[np.random.randint(len(labelmarkimg))]
showimage_shape = showimage.shape
# use gray scale color map if there is only one channel
if len(showimage_shape) == 2:
cmap = "gray"
elif showimage_shape[2] == 1:
showimage = showimage[:,:,0]
cmap = "gray"
plt.imshow(showimage,cmap=cmap)
plt.title(i,fontsize=50,fontweight='bold')
plt.xticks([])
plt.yticks([])
plt.tight_layout(pad=0, h_pad=0, w_pad=0)
plt.show()
def show_class_by_class_images(images, lables, classes=n_classes, cols = 10,cmap=None):
""""
Show all class by class from dataset.
Arguments:
images: source images
lables: source corresponding label
sumclass: sumary class(42)
"""
for i in range(classes):
show_one_class_images(images,lables,i,cols)
Random show one specified class image.¶
The first element is an exmaple of this specified sign.
In [4]:
show_one_class_images(X_train,y_train,21)
Random show one image of all class from dataset.¶
In [5]:
show_all_class_images(X_train,y_train)
# show_all_class_images(X_valid,y_valid)
# show_all_class_images(X_test,y_test)
Show all class by class from dataset.¶
In [6]:
show_class_by_class_images(X_train,y_train)
Histogram visualization function defintion¶
In [7]:
# histogram of label frequency
# show each class
def show_single_class_histogram(trainlabel, sumclass=n_classes):
""""
show single images histogram
Arguments:
images: source images
lables: source corresponding label
"""
hist_train, bins_train = np.histogram(trainlabel, bins=sumclass)
width = 0.8 * (bins_train[1] - bins_train[0])
#adjust center location 1.96
center = (bins_train[:-1] + bins_train[1:]) / 1.96
plt.figure(figsize=(15,5))
l1 = plt.bar(center-width/2, hist_train, align='center', width=width)
plt.xticks(np.arange(0, sumclass, 1) )
plt.xlabel('Classes')
plt.ylabel('Count of each class')
plt.show()
def show_compared_class_histogram(trainlabel, testlabel, sumclass=n_classes):
""""
Show compared class histogram
Arguments:
images: source images
lables: source corresponding label
"""
hist_train, bins_train = np.histogram(trainlabel, bins=sumclass)
hist_test, bins_test = np.histogram(testlabel, bins=sumclass)
width = 0.5 * (bins_train[1] - bins_train[0])
#adjust center location 1.96
center = (bins_train[:-1] + bins_train[1:]) / 1.96
plt.figure(figsize=(15,5))
l1 = plt.bar(center-width, hist_train, align='center', width=width)
l2 = plt.bar(center, hist_test, align='center', width=width)
plt.legend((l1, l2), ('first set', 'second set'))
plt.xticks(np.arange(0, sumclass, 1) )
plt.xlabel('Classes')
plt.ylabel('Numbers of each class')
plt.show()
Histogram visualization of the original train,valid and test dataset¶
In [8]:
print("Train dataset showing:")
show_single_class_histogram(y_train)
print("Valid dataset showing:")
show_single_class_histogram(y_valid)
print("Test dataset showing:")
show_single_class_histogram(y_test)
Histogram visualization comparation of the original train and test dataset¶
In [9]:
print("Train dataset compared with test dataset showing:")
show_compared_class_histogram(y_train,y_test,n_classes)
Step 2: Design and Test a Model Architecture¶
Design and implement a deep learning model that learns to recognize traffic signs.
Those aspects will be considered in this chapter:
- Neural network architecture (is the network over or underfitting?)
- Play around preprocessing techniques (normalization, rgb to grayscale, etc)
- Number of examples per label (some have more than others).
- Generate fake data to banlance each class.
Generate fake data to balance image numbers of per label¶
Image Data Generator function definition¶
In [10]:
import cv2
from skimage import io, color, exposure, transform
from skimage.util import random_noise
def gamma_correction(img, correction):
""""
Gamma correction function
"""
img = img/255.0
img = cv2.pow(img, correction)
return np.uint8(img*255)
def gammaImage(img,gammarate= 0.4):
""""
Gamma correction function as expection of brightness lighting conditions
"""
HSV = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)
# HSV[:,:,2] = gamma_correction((HSV[:,:,2]),0.6)
HSV[:,:,2] = gamma_correction((HSV[:,:,2]),(2*gammarate * np.random.random_sample(1) + 1-gammarate))
img = cv2.cvtColor(HSV, cv2.COLOR_HSV2RGB )
return img
def gaussian_blur(img, kernel_size=3):
""""
Gaussian blur function as expection of motion-blur
"""
return cv2.GaussianBlur(img, (kernel_size, kernel_size), 0)
def gaussian_noise(img):
""""
Gaussian noise function as expection of noise
"""
var = 0.001*np.random.random_sample(1)
return np.uint8(255*random_noise(img/255, mode="gaussian",var=var))
def randomerase(img, hlimt=14,wlimit=14):
""""
Gaussian erase function as expection of occlusions or physical damage
"""
imgshape = img.shape
h = np.random.randint(0, hlimt)
w = np.random.randint(0, wlimit)
x1 = np.random.randint(0, imgshape[0] - h)
y1 = np.random.randint(0, imgshape[1] - w)
cropimg =np.copy(img)
cropimg[x1:x1+h, y1:y1+w,0] = 127
cropimg[x1:x1+h, y1:y1+w,1] = 127
cropimg[x1:x1+h, y1:y1+w,2] = 127
return cropimg
def DataAugumentProcess(img,gammarate = 0.5,blurrate = 0.2,randomeraserte=0.2):
""""
Data Augumention Process function
"""
img = gammaImage(img,gammarate)
if(np.random.random_sample(1) < blurrate):
img = gaussian_blur(img)
if(np.random.random_sample(1) < randomeraserte):
img = randomerase(img)
img = gaussian_noise(img)
return img
Image Data Generator function definition to show¶
In [11]:
# creat the generator to perform online data augmentation
from keras.preprocessing.image import ImageDataGenerator
image_datagen = ImageDataGenerator(rotation_range=5.,
shear_range=0.2,
zoom_range=0.2,
channel_shift_range=25,
width_shift_range=0.1,
height_shift_range=0.1,
preprocessing_function=DataAugumentProcess)
# take a random image from the training set
def gen_class_images_show(images,lables,class_n):
""""
Generate specified one class image for show testing
"""
imagemarkimg = np.copy(X_train[y_train == class_n])
labelmarkimg = y_train[y_train == class_n]
img_rgb = imagemarkimg[np.random.randint(len(labelmarkimg))].astype(np.uint8)
# plot the original image
plt.figure(figsize=(2,2))
plt.imshow(img_rgb)
plt.title('Source of RGB image (class = {})'.format(class_n))
plt.axis('off')
plt.show()
x = np.empty((0,img_rgb.shape[0],img_rgb.shape[1],img_rgb.shape[2]),dtype='uint8')
y = np.empty(0,dtype='uint8')
print('Generating of RGB image (class = {})'.format(class_n))
print()
batch_size = 30
for x_batch, y_batch in image_datagen.flow(imagemarkimg,labelmarkimg, batch_size=batch_size):
# print(x.shape)
# print(x_batch.shape)
x = np.concatenate((x,x_batch.astype(np.uint8)),axis=0)
y = np.concatenate((y,y_batch),axis=0)
# print(x.shape)
break
show_one_class_images(x,y,class_n,batch_size,False)
Image Data Generator function show test¶
In [12]:
gen_class_images_show(X_train,y_train,21)
Image Data Generator function definition¶
In [13]:
def gen_class_images(images,lables,sumclass,n_min_samples = 6000,batch_size = 128):
""""
Generate specified one class image
"""
for class_n in range(0,sumclass):
print("Class ",class_n, ': ', end='')
imagemarkimg = images[lables == class_n]
labelmarkimg = lables[lables == class_n]
n_samples = len(imagemarkimg)
print(n_samples,end='')
batch_counter = (n_min_samples-n_samples)
while( batch_counter >= 0):
for x_batch, y_batch in image_datagen.flow(imagemarkimg, labelmarkimg,batch_size):
images = np.concatenate((images, x_batch.astype(np.uint8)), axis=0)
lables = np.concatenate((lables, y_batch), axis=0)
if batch_counter < 0:
break
batch_counter -= batch_size
n_samples = len(images[lables == class_n])
batch_counter = (n_min_samples-n_samples)
print(' Generating samples numbers = {}'.format(n_samples))
return images,lables
Train dataset enlarge¶
In [16]:
X_train = np.concatenate((X_train, X_valid), axis=0)
y_train = np.concatenate((y_train, y_valid), axis=0)
Image Data Generating¶
In [17]:
X_train,y_train = gen_class_images(X_train,y_train,n_classes)
Generated Data Save¶
In [18]:
train_dataset_file = "traffic-signs-data/gen_train.p"
print("Number of training examples after augmenting and balancing training data =", X_train.shape[0])
pickle.dump({
"features" : X_train,
"labels" : y_train
}, open(train_dataset_file, "wb" ),protocol=4)
print("Generated balanced dataset saved in", train_dataset_file)
Generated Data Restored¶
In [14]:
train_dataset_file = "traffic-signs-data/gen_train.p"
with open(train_dataset_file, mode='rb') as f:
gen_train = pickle.load(f)
X_train, y_train = gen_train['features'], gen_train['labels']
print("Number of training examples after augmenting and balancing training data =", X_train.shape[0])
Generate train and valid dateset¶
In [15]:
from sklearn.model_selection import train_test_split
X_train, X_valid, y_train, y_valid = train_test_split(X_train, y_train, test_size=0.2, random_state=42)
show_compared_class_histogram(y_train,y_valid,n_classes)
Pre-process the Data Set (normalization, grayscale, etc.)¶
Pre-process dataset from 0~255 to -0.5~0.5.
In [16]:
def normalizeyImage(img):
# Histogram normalization in y
# return cv2.cvtColor(img, cv2.COLOR_YCrCb2GRAY)
# return np.resize(cv2.equalizeHist(cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)),(32,32,1))
img = img/255.0-0.5
return (img)
def preprocessImages(images):
ret_array = []
for img in images:
ret_array.append(normalizeyImage((img)))
return np.array(ret_array)
X_train_pre = preprocessImages(X_train)
X_valid_pre = preprocessImages(X_valid)
X_test_pre = preprocessImages(X_test)
Model Architecture¶
In [17]:
### Define your architecture here.
### Feel free to use as many code cells as needed.
import tensorflow as tf
from tensorflow.contrib.layers import flatten
from sklearn.utils import shuffle
def VGGlike(x):
# Arguments used for tf.truncated_normal, randomly defines variables for the weights and biases for each layer
mu = 0
sigma = 0.05
keep_prob1 = tf.placeholder(tf.float32)
keep_prob2 = tf.placeholder(tf.float32)
keep_prob3 = tf.placeholder(tf.float32)
phase = tf.placeholder(tf.bool)
# SOLUTION: Layer 1: Convolutional. Input = 32x32x3. Output = 32x32x32
conv1_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 3, 32), mean = mu, stddev = sigma))
conv1_b = tf.Variable(tf.zeros(32))
conv1 = tf.nn.conv2d(x, conv1_W, strides=[1, 1, 1, 1], padding='SAME') + conv1_b
# SOLUTION: Activation.
conv1 = tf.nn.relu(conv1,name="conv1")
# conv1 = tf.nn.dropout(conv1, keep_prob1)
# SOLUTION: Layer 2: Convolutional. Input = 32x32x32. Output = 32x32x32
conv2_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 32, 32), mean = mu, stddev = sigma))
conv2_b = tf.Variable(tf.zeros(32))
conv2 = tf.nn.conv2d(conv1, conv2_W, strides=[1, 1, 1, 1], padding='SAME') + conv2_b
# SOLUTION: Activation.
conv2 = tf.nn.relu(conv2,name="conv2")
# SOLUTION: Pooling. Input = 32x32x32. Output = 16x16x32.
conv2 = tf.nn.max_pool(conv2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
conv2 = tf.nn.dropout(conv2, keep_prob1)
# SOLUTION: Layer 3: Convolutional. Input = 16x16x32. Output = 16x16x64
conv3_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 32, 64), mean = mu, stddev = sigma))
conv3_b = tf.Variable(tf.zeros(64))
conv3 = tf.nn.conv2d(conv2, conv3_W, strides=[1, 1, 1, 1], padding='SAME') + conv3_b
# SOLUTION: Activation.
conv3 = tf.nn.relu(conv3,name="conv3")
# conv3 = tf.nn.dropout(conv3, keep_prob1)
# SOLUTION: Layer 4: Convolutional. Input = 16x16x64. Output = 16x16x64
conv4_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 64, 64), mean = mu, stddev = sigma,name="conv4_W"))
conv4_b = tf.Variable(tf.zeros(64))
conv4 = tf.nn.conv2d(conv3, conv4_W, strides=[1, 1, 1, 1], padding='SAME') + conv4_b
# SOLUTION: Activation.
conv4 = tf.nn.relu(conv4,name="conv4")
# SOLUTION: Pooling. Input = 10x10x16. Output = 5x5x16.
conv4 = tf.nn.max_pool(conv4, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
conv4 = tf.nn.dropout(conv4, keep_prob1)
# SOLUTION: Layer 5: Convolutional. Input = 16x16x64. Output = 16x16x128
conv5_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 64, 128), mean = mu, stddev = sigma,name="conv5_W"))
conv5_b = tf.Variable(tf.zeros(128))
conv5 = tf.nn.conv2d(conv4, conv5_W, strides=[1, 1, 1, 1], padding='SAME') + conv5_b
# SOLUTION: Activation.
conv5 = tf.nn.relu(conv5,name="conv5")
# conv5 = tf.nn.dropout(conv5, keep_prob2)
# SOLUTION: Layer 6: Convolutional. Input = 16x16x128. Output = 16x16x128
conv6_W = tf.Variable(tf.truncated_normal(shape=(3, 3, 128, 128), mean = mu, stddev = sigma,name="conv6_W"))
conv6_b = tf.Variable(tf.zeros(128))
conv6 = tf.nn.conv2d(conv5, conv6_W, strides=[1, 1, 1, 1], padding='SAME') + conv6_b
# SOLUTION: Activation.
conv6 = tf.nn.relu(conv6,name="conv6")
# SOLUTION: Pooling. Input = 16x16x128. Output = 8x8x128.
conv6 = tf.nn.max_pool(conv6, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='VALID')
conv6 = tf.nn.dropout(conv6, keep_prob2)
fc0 = flatten(conv6)
# SOLUTION: Layer 1: Fully Connected. Input = 2048. Output = 2048.
fc1_W = tf.Variable(tf.truncated_normal(shape=(2048, 2048), mean = mu, stddev = sigma))
fc1_b = tf.Variable(tf.zeros(2048))
fc1 = tf.matmul(fc0, fc1_W) + fc1_b
# SOLUTION: Activation.
fc1 = tf.nn.relu(fc1)
fc1 = tf.nn.dropout(fc1, keep_prob3)
# SOLUTION: Layer 2: Fully Connected. Input = 2048. Output = 2048.
fc2_W = tf.Variable(tf.truncated_normal(shape=(2048, 2048), mean = mu, stddev = sigma))
fc2_b = tf.Variable(tf.zeros(2048))
fc2 = tf.matmul(fc1, fc2_W) + fc2_b
# SOLUTION: Activation.
fc2 = tf.nn.relu(fc2)
fc2 = tf.nn.dropout(fc2, keep_prob3)
# SOLUTION: Layer 3: Fully Connected. Input = 2048. Output = 43.
fc3_W = tf.Variable(tf.truncated_normal(shape=(2048, 43), mean = mu, stddev = sigma))
fc3_b = tf.Variable(tf.zeros(43))
logits = tf.matmul(fc2, fc3_W) + fc3_b
regularizers = tf.nn.l2_loss(fc1_W) + tf.nn.l2_loss(fc2_W) + tf.nn.l2_loss(fc3_W)
# regularizers = tf.nn.l2_loss(conv1_W) + tf.nn.l2_loss(conv2_W) + tf.nn.l2_loss(conv3_W) +\
# tf.nn.l2_loss(conv4_W) + tf.nn.l2_loss(conv5_W) + tf.nn.l2_loss(conv6_W) + tf.nn.l2_loss(fc1_W) + tf.nn.l2_loss(fc2_W)+tf.nn.l2_loss(fc3_W)
return logits,keep_prob1,keep_prob2,keep_prob3,phase,regularizers
Model Train Function Definition¶
In [18]:
# rate = 0.001
# EPOCHS = 50
# BATCH_SIZE = 128
x = tf.placeholder(tf.float32, (None, 32, 32, 3))
y = tf.placeholder(tf.int32, (None))
one_hot_y = tf.one_hot(y, 43)
logits,keep_prob1,keep_prob2,keep_prob3,phase,loss_regularizers = VGGlike(x)
cross_entropy = tf.nn.softmax_cross_entropy_with_logits(labels=one_hot_y, logits=logits)
# loss_operation = tf.reduce_mean(cross_entropy+ 0.0005*loss_regularizers)
# optimizer = tf.train.AdamOptimizer(learning_rate = rate)
# training_operation = optimizer.minimize(loss_operation)
correct_prediction = tf.equal(tf.argmax(logits, 1), tf.argmax(one_hot_y, 1))
accuracy_operation = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
saver = tf.train.Saver()
def evaluate(X_data, y_data,BATCH_SIZE):
num_examples = len(X_data)
total_accuracy = 0
sess = tf.get_default_session()
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
accuracy = sess.run(accuracy_operation, feed_dict={x: batch_x, y: batch_y,keep_prob1: 1.0,keep_prob2: 1.0,keep_prob3: 1.0,phase: 0})
total_accuracy += (accuracy * len(batch_x))
return total_accuracy / num_examples
Train, Validate and Test the Model¶
A validation set can be used to assess how well the model is performing. A low accuracy on the training and validation sets imply underfitting. A high accuracy on the training set but low accuracy on the validation set implies overfitting.
Train model¶
In [35]:
### Train your model here.
### Calculate and report the accuracy on the training and validation set.
### Once a final model architecture is selected,
### the accuracy on the test set should be calculated and reported as well.
### Feel free to use as many code cells as needed.
rate = 0.0005
lossrate = 0.0001
loss_operation = tf.reduce_mean(cross_entropy + lossrate*loss_regularizers)
optimizer = tf.train.AdamOptimizer(learning_rate = rate)
training_operation = optimizer.minimize(loss_operation)
def trainmodel(x_train_data,y_train_data,x_valid_data,y_valid_data,EPOCHS = 80,BATCH_SIZE = 128,logstart = 0.3,require_improvement = 20):
# Best validation accuracy seen so far.
best_validation_accuracy = 0.0
start_log = EPOCHS * logstart
# Iteration-number for last improvement to validation accuracy.
last_improvement = start_log
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
num_examples = len(x_train_data)
print("Training...")
print()
for i in range(EPOCHS):
x_train_data, y_train_data = shuffle(x_train_data, y_train_data)
for offset in range(0, num_examples, BATCH_SIZE):
end = offset + BATCH_SIZE
batch_x, batch_y = x_train_data[offset:end], y_train_data[offset:end]
sess.run(training_operation, feed_dict={x: batch_x, y: batch_y,keep_prob1: 0.9,keep_prob2: 0.8,keep_prob3:0.5,phase: 1})
test_accuracy = evaluate(x_train_data, y_train_data,BATCH_SIZE)
print("EPOCH {} ...".format(i+1),end=' ')
print("test Accuracy = {:.4f}".format(test_accuracy),end=' ')
validation_accuracy = evaluate(x_valid_data, y_valid_data,BATCH_SIZE)
print("Validation Accuracy = {:.4f}".format(validation_accuracy))
# If validation accuracy is an improvement over best-known.
if i > start_log and validation_accuracy > best_validation_accuracy:
# Update the best-known validation accuracy.
best_validation_accuracy = validation_accuracy
# Set the iteration for the last improvement to current.
last_improvement = i
# Save all variables of the TensorFlow graph to file.
saver.save(sess, './lenet')
# If no improvement found in the required number of iterations.
if i - last_improvement > require_improvement:
print("No improvement found in a loop, stopping optimization.")
# Break out from the for-loop.
break
# saver.save(sess, './lenet')
# print("Model saved")
# Print result.
print("best Accuracy = {:.4f} Model saved".format(best_validation_accuracy))
In [36]:
trainmodel(X_train_pre,y_train,X_valid_pre,y_valid)
Finetune model¶
In [44]:
rate = 0.0001
lossrate = 0.0001
loss_operation = tf.reduce_mean(cross_entropy + lossrate*loss_regularizers)
optimizer = tf.train.AdamOptimizer(learning_rate = rate)
training_operation = optimizer.minimize(loss_operation)
def finetunemodel(x_train_data,y_train_data,x_valid_data,y_valid_data,EPOCHS = 80,BATCH_SIZE = 128,logstart = 0,require_improvement = 10):
# Best validation accuracy seen so far.
best_validation_accuracy = 0.0
start_log = EPOCHS * logstart
# Iteration-number for last improvement to validation accuracy.
last_improvement = start_log
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
saver.restore(sess, "./lenet")
num_examples = len(x_train_data)
print("Finetune Training...")
print()
for i in range(EPOCHS):
x_train_data, y_train_data = shuffle(x_train_data, y_train_data)
for offset in range(0, num_examples, BATCH_SIZE):
end = offset + BATCH_SIZE
batch_x, batch_y = x_train_data[offset:end], y_train_data[offset:end]
sess.run(training_operation, feed_dict={x: batch_x, y: batch_y,keep_prob1: 0.9,keep_prob2: 0.8,keep_prob3:0.5,phase: 1})
test_accuracy = evaluate(x_train_data, y_train_data,BATCH_SIZE)
print("EPOCH {} ...".format(i+1),end=' ')
print("test Accuracy = {:.4f}".format(test_accuracy),end=' ')
validation_accuracy = evaluate(x_valid_data, y_valid_data,BATCH_SIZE)
print("Validation Accuracy = {:.4f}".format(validation_accuracy))
# If validation accuracy is an improvement over best-known.
if i > start_log and validation_accuracy > best_validation_accuracy:
# Update the best-known validation accuracy.
best_validation_accuracy = validation_accuracy
# Set the iteration for the last improvement to current.
last_improvement = i
# Save all variables of the TensorFlow graph to file.
saver.save(sess, './lenet')
print("Model Save!")
# If no improvement found in the required number of iterations.
if i - last_improvement > require_improvement:
print("No improvement found in a loop, stopping optimization.")
# Break out from the for-loop.
break
# Print result.
print("best Accuracy = {:.4f} Model saved".format(best_validation_accuracy))
In [45]:
finetunemodel(X_train_pre,y_train,X_valid_pre,y_valid)
Evaluate on test dataset¶
In [19]:
def testdataset(X_data, y_data,BATCH_SIZE = 128):
with tf.Session() as sess:
# saver.restore(sess, tf.train.latest_checkpoint('.'))
saver.restore(sess, "./lenet")
test_accuracy = evaluate(X_data, y_data,BATCH_SIZE)
print("Test Accuracy = {:.4f}".format(test_accuracy))
In [20]:
testdataset(X_test_pre,y_test)
Trained Model Summary¶
Designed model get 99.99% on traindataset 99.06% on validdataset 99.36% on test dataset.
Analyze Model Performance¶
In [21]:
def predict(X_data, y_data,fun,BATCH_SIZE = 128):
num_examples = len(y_data)
total_accuracy = 0
ret_array = []
sess = tf.get_default_session()
for offset in range(0, num_examples, BATCH_SIZE):
batch_x, batch_y = X_data[offset:offset+BATCH_SIZE], y_data[offset:offset+BATCH_SIZE]
lable = sess.run(fun, feed_dict={x: batch_x, y: batch_y,keep_prob1: 1.0,keep_prob2: 1.0,keep_prob3: 1.0,phase: 0})
ret_array.extend(lable)
return np.array(ret_array)
In [22]:
def labelpredict(X_data, y_data,logits=logits,BATCH_SIZE=128):
y_pred = tf.nn.softmax(logits=logits)
y_pred_cls = tf.argmax(y_pred, axis=1)
lables = []
with tf.Session() as sess:
saver.restore(sess, tf.train.latest_checkpoint('.'))
lables = predict(X_data, y_data,y_pred_cls)
return lables
In [23]:
preditlables = labelpredict(X_test_pre,y_test)
In [24]:
def show_all_images(images,lables,cmap=None):
cols = 15
sumclass = len(images)
rows = (sumclass+cols-1)//cols
plt.figure(figsize=(30, 20))
for i in range(sumclass):
plt.subplot(rows, cols, i+1)
showimage = images[i]
# use gray scale color map if there is only one channel
showimage_shape = showimage.shape
if len(showimage_shape) == 2:
cmap = "gray"
elif showimage_shape[2] == 1:
showimage = showimage[:,:,0]
cmap = "gray"
plt.imshow(showimage,cmap=cmap)
plt.title(lables[i],fontsize=30,fontweight='bold')
plt.xticks([])
plt.yticks([])
plt.tight_layout(pad=0, h_pad=0, w_pad=0)
plt.show()
In [25]:
incorrectly_predictedx = X_test[y_test != preditlables]
incorrectly_predictedy = y_test[y_test != preditlables]
print("Total {} incorrectly predicted images :".format(len(incorrectly_predictedy)))
show_all_images(incorrectly_predictedx,incorrectly_predictedy)
In [26]:
print("incorrectly predicted one specified images summary:")
show_one_class_images(incorrectly_predictedx,incorrectly_predictedy,12,50,False)
In [27]:
print("histogram of incorrectly predited labels frequency:")
show_single_class_histogram(incorrectly_predictedy,n_classes)
Step 3: Test a Model on New Images¶
To give yourself more insight into how your model is working, download at least five pictures of German traffic signs from the web and use your model to predict the traffic sign type.
You may find signnames.csv useful as it contains mappings from the class id (integer) to the actual sign name.
Load and Output the Images¶
In [28]:
### Load the images and plot them here.
### Feel free to use as many code cells as needed.
from skimage import transform,data
import glob
def show_images(images, cmap=None):
cols = 5
rows = (len(images)+2)//cols
plt.figure(figsize=(8, 8))
for i, image in enumerate(images):
plt.subplot(rows, cols, i+1)
# use gray scale color map if there is only one channel
cmap = 'gray' if len(image.shape)==2 else cmap
plt.imshow(image, cmap=cmap)
plt.xticks([])
plt.yticks([])
plt.tight_layout(pad=0, h_pad=0, w_pad=0)
plt.show()
test_images = [plt.imread(path) for path in glob.glob('test_images/*.jpg')]
test_images = np.reshape(test_images,(len(test_images),)+ test_images[0].shape)
show_images(test_images)
Predict the Sign Type for Each Image¶
In [29]:
### Run the predictions here and use the model to output the prediction for each image.
### Make sure to pre-process the images with the same pre-processing pipeline used earlier.
### Feel free to use as many code cells as needed.
#Visualize new raw images
def test_net(X_data, sess):
prob = sess.run(tf.nn.softmax(logits), feed_dict={x: X_data,keep_prob1: 1.0,keep_prob2: 1.0,keep_prob3: 1.0, phase: 0})
top_5 = tf.nn.top_k(prob, k=5)
return sess.run(top_5)
with tf.Session() as sess:
saver.restore(sess, "./lenet")
test_images_pre = preprocessImages(test_images)
signs_top_5=test_net(test_images_pre,sess)
print("Predition on new images:")
plt.figure(figsize=(12, 12))
for i in range(5):
plt.subplot(5, 3, 3*i+1)
plt.imshow(test_images[i])
# plt.title(i)
plt.axis('off')
plt.subplot(5, 3, 3*i+3)
plt.barh(np.arange(1, 6, 1), signs_top_5.values[i, :])
labs=[signnames[j] for j in signs_top_5.indices[i]]
plt.yticks(np.arange(1, 6, 1), labs)
plt.show()
Analyze Performance¶
Calculate the accuracy for these 5 new images.¶
The model predicted 5 of 5 signs correctly, it's 100% accurate on these new images.
Output Top 5 Softmax Probabilities For Each Image Found on the Web¶
For each of the new images, print out the model's softmax probabilities to show the certainty of the model's predictions (limit the output to the top 5 probabilities for each image). tf.nn.top_k could prove helpful here.
The example below demonstrates how tf.nn.top_k can be used to find the top k predictions for each image.
tf.nn.top_k will return the values and indices (class ids) of the top k predictions. So if k=3, for each sign, it'll return the 3 largest probabilities (out of a possible 43) and the correspoding class ids.
Take this numpy array as an example. The values in the array represent predictions. The array contains softmax probabilities for five candidate images with six possible classes. tf.nn.top_k is used to choose the three classes with the highest probability:
# (5, 6) array
a = np.array([[ 0.24879643, 0.07032244, 0.12641572, 0.34763842, 0.07893497,
0.12789202],
[ 0.28086119, 0.27569815, 0.08594638, 0.0178669 , 0.18063401,
0.15899337],
[ 0.26076848, 0.23664738, 0.08020603, 0.07001922, 0.1134371 ,
0.23892179],
[ 0.11943333, 0.29198961, 0.02605103, 0.26234032, 0.1351348 ,
0.16505091],
[ 0.09561176, 0.34396535, 0.0643941 , 0.16240774, 0.24206137,
0.09155967]])Running it through sess.run(tf.nn.top_k(tf.constant(a), k=3)) produces:
TopKV2(values=array([[ 0.34763842, 0.24879643, 0.12789202],
[ 0.28086119, 0.27569815, 0.18063401],
[ 0.26076848, 0.23892179, 0.23664738],
[ 0.29198961, 0.26234032, 0.16505091],
[ 0.34396535, 0.24206137, 0.16240774]]), indices=array([[3, 0, 5],
[0, 1, 4],
[0, 5, 1],
[1, 3, 5],
[1, 4, 3]], dtype=int32))Looking just at the first row we get [ 0.34763842, 0.24879643, 0.12789202], you can confirm these are the 3 largest probabilities in a. You'll also notice [3, 0, 5] are the corresponding indices.
In [30]:
### Print out the top five softmax probabilities for the predictions on the German traffic sign images found on the web.
### Feel free to use as many code cells as needed.
print(signs_top_5)
Step 4 (Optional): Visualize the Neural Network's State with Test Images¶
This Section is not required to complete but acts as an additional excersise for understaning the output of a neural network's weights. While neural networks can be a great learning device they are often referred to as a black box. We can understand what the weights of a neural network look like better by plotting their feature maps. After successfully training your neural network you can see what it's feature maps look like by plotting the output of the network's weight layers in response to a test stimuli image. From these plotted feature maps, it's possible to see what characteristics of an image the network finds interesting. For a sign, maybe the inner network feature maps react with high activation to the sign's boundary outline or to the contrast in the sign's painted symbol.
Your output should look something like this (above)
In [31]:
### Visualize your network's feature maps here.
### Feel free to use as many code cells as needed.
# image_input: the test image being fed into the network to produce the feature maps
# tf_activation: should be a tf variable name used during your training procedure that represents the calculated state of a specific weight layer
# activation_min/max: can be used to view the activation contrast in more detail, by default matplot sets min and max to the actual min and max values of the output
# plt_num: used to plot out multiple different weight feature map sets on the same block, just extend the plt number for each new feature map entry
def outputFeatureMap(sess,image_input, tf_activation, activation_min=-1, activation_max=-1 ,plt_num=1):
# Here make sure to preprocess your image_input in a way your network expects
# with size, normalization, ect if needed
# image_input =
# Note: x should be the same name as your network's tensorflow data placeholder variable
# If you get an error tf_activation is not defined it may be having trouble accessing the variable from inside a function
activation = tf_activation.eval(session=sess,feed_dict={x : image_input,keep_prob1: 1.0,keep_prob2: 1.0,keep_prob3: 1.0,phase: 0})
print("layer shape is {}".format(activation.shape))
featuremaps = activation.shape[3]
plt.figure(plt_num, figsize=(15,15))
if featuremaps > 48:
featuremaps = 48
for featuremap in range(featuremaps):
plt.subplot(6,8, featuremap+1) # sets the number of feature maps to show on each row and column
plt.title('FeatureMap ' + str(featuremap)) # displays the feature map number
if activation_min != -1 & activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin =activation_min, vmax=activation_max, cmap="gray")
elif activation_max != -1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmax=activation_max, cmap="gray")
elif activation_min !=-1:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", vmin=activation_min, cmap="gray")
else:
plt.imshow(activation[0,:,:, featuremap], interpolation="nearest", cmap="gray")
Output Each Layer¶
In [32]:
def outputFeatureMapLayer(image_input,layer):
with tf.Session() as sess:
saver.restore(sess, tf.train.latest_checkpoint('.'))
conv_layer = tf.get_default_graph().get_tensor_by_name("conv{}:0".format(layer))
print("conv layer is:")
print(conv_layer)
outputFeatureMap(sess,image_input,conv_layer)
In [33]:
img_rgb_origin = X_test[1]
img_rgb = X_test_pre[1]
image_input = np.reshape(img_rgb,(1,)+ img_rgb.shape)
print("input image shape is {}".format(image_input.shape))
plt.figure(figsize=(2, 2))
plt.imshow(img_rgb_origin)
plt.axis('off')
plt.show()
conv1 Layer¶
In [34]:
outputFeatureMapLayer(image_input,1)
conv2 Layer¶
In [35]:
outputFeatureMapLayer(image_input,2)
conv3 Layer¶
In [36]:
outputFeatureMapLayer(image_input,3)
conv4 Layer¶
In [37]:
outputFeatureMapLayer(image_input,4)
conv5 Layer¶
In [38]:
outputFeatureMapLayer(image_input,5)
conv6 Layer¶
In [39]:
outputFeatureMapLayer(image_input,6)
